AI agent architecture
An agent that can call tools is useful only if you can list the tools, the policy, and the eval that proves it stayed inside both.
Pieces
- Planner / loop (usually a single model with a tool list).
- Tools with least privilege.
- Approval for irreversible actions.
- Short memory or none—long memory is a product decision with privacy cost.
- Traces for replay.
Agent development · What is an AI agent? · Agent vs chatbot
If the problem maps to work we actually ship, we will say so in 20 minutes.
Request a fit call